uditgoenka/autoresearch v2.2.3

on GitHub

What's new in v2.2.3

Guard hooks only run during an autoresearch run (#125, #123)

• scout-block, dangerous-cmd-block, simplify-gate and dev-rules-reminder no longer fire in sessions that are not running autoresearch.
• A run is active after you type /autoresearch, when the model loads the autoresearch skill, or while a results TSV was written in the last 30 minutes.
• The run flag now survives compact and resume.
• privacy-block stays global.

Fewer false blocks

• git push --force-with-lease and --force-if-includes are allowed (#125). Plain --force, -f and +refspec pushes are still blocked.
• simplify-gate skips untracked binary files (#119).

Fewer tokens per turn

• Subcommands drop out of the model's per-turn skill listing (about 718 tokens per turn). You can still type them, and the model reaches them through SKILL.md.

Fixes

• Plateau history is read from orchestrator state (#127, thanks @chupakobra6).

Update

• Claude Code: claude plugin marketplace update autoresearch && claude plugin update autoresearch@autoresearch
• Codex: npx skills update

Full change: #128